99% Availability

The Downtime Math

 AVAILABILITY = uptime ÷ total time

 99% means 1 failure minute allowed per 100:

 per YEAR    = 525,600 min × 1%   ≈ 3.65 days down
 per MONTH   = 43,800 min × 1%    ≈ 7.3 hours down
 per WEEK    = 10,080 min × 1%    ≈ 1.7 hours down
 per DAY     = 1,440 min × 1%     ≈ 14.4 minutes down

Read that yearly number again: three and a half days. Two nines is a generous budget, not a demanding one.

What 99% Permits Architecturally

A budget this large changes engineering options:

• Single-AZ deployment is viable: one datacenter’s occasional maintenance fits the budget.
• Business-hours deploys: restart during low traffic; no zero-downtime machinery required.
• Simple failover: manual promotion of a standby is acceptable when it costs 20 minutes twice a year.
• Single region, single provider: no multi-cloud hedging.

Every one of these saves real money and complexity — which is why 99% remains a deliberate choice for many systems.

Who Genuinely Runs at Two Nines

System class	Why it’s acceptable
Internal admin tools	Staff tolerate scheduled windows
Batch/analytics platforms	Consumers are pipelines with retry logic
Pre-launch products	Velocity > availability; rebuild later
Regional B2B tools with contracts permitting windows	Maintenance clauses exist

The anti-pattern is accidental 99%: consumer products whose teams never chose the number and discover it during an outage tweet-storm.

The Budget Mindset

Availability targets convert to spending decisions:

 error budget (SRE framing): 1% of minutes = your monthly "down allowance"
 
 spent on: deploys, experiments, incidents, maintenance
 exhausted → feature work pauses; reliability work resumes
 
 at 99%, the budget funds aggressive shipping cadence —
 a genuine product advantage over a trembling 5-nines fleet

Choosing Honestly

The selection question is user-experience arithmetic:

 "how bad is 14 minutes of downtime TODAY for our users?"
 - internal dashboard:      mild annoyance        → 99% fine
 - food delivery at lunch:  revenue + trust hit   → need more nines
 - payment capture:         contractual penalties → far more nines

Match the tier to consequence, not ambition.

Interview Framing

When interviews ask “what availability do you design for?”, the weak answer is reflexive maximalism (“five nines!”). Strong answers reason from consequence: “this is a consumer ordering flow — I’d take 99.9% for browsing but treat payment capture as its own stricter tier.” Showing you know what 99% permits and when it’s wrong demonstrates calibration, not laziness.